mangotee
commented
4 years ago Hi Varshul, thanks for your interest and for the positive feedback. What do you mean "low-dimensional" data? DeepVOG input images are supposed to have a resolution of 320x240. Would your images have a lower resolution? We have not tested pupil detection with lower resolutions, however, DeepVOG was trained on images that are relatively blurry, so a high-quality upsampling (e.g. bicubic, or deep supersampling ;) ) of your source material might still lead to successful detection. Regarding dataset - I fully agree, sharing data is really beneficial. However, we have to make sure that we are able and allowed to do so. I will investigate and post an update once I know. Until then, the pre-trained model is available in the models folder. You can probably use very little data (e.g. 50-100 labeled images, max 500) to fine-tune the segmentation on a representative dataset from your domain. If you need a labeling tool, let us know, we could include our custom tool that we used for labeling (very simple, matplotlib-based) into our repo in the meantime as well.
szymon319
yyhhoi
guardexcorp
Hey,
Good work. We are looking to similar lines but with really low dimensional data. We are currently getting data from Webcam.
Can we please have dataset links as well? We and the community can greatly benefit from the same.